Conrad Watt

I'd prefer the second option - I also think a name change like "functional-determinism" or "numeric-determinism" would better reflect the scope of such a profile, but I wouldn't push for...

I'd also like to attend tomorrow to discuss the non-determinism point that @penzn mentioned in advance of the phase 4 vote. I've just filled out the form linked above. EDIT:...

I'll post a better summation comment later, but for now here's a [raw link to my slides](https://docs.google.com/presentation/d/1dJhqnAZOjh8ppCKivDl--niaUgpKV44--IMX1EF1SA0/edit?usp=sharing).

This comment is to sum up my position and attempt to distill some conversations and points that were raised at the previous meeting. I'm focussing on the `fma` discussion here...

> My understanding (and @Maratyszcza, please correct me if I'm wrong) is that in all cases there is a better option than emulating single-rounding vector `fma`, so no program would...

> something about different code in the tools that would target which FMA instruction to generate, perhaps I missed or misunderstood something? For a toolchain currently built around option (a),...

> I will note that currently no toolchain reasons about any of this and that all relaxed SIMD usage (that I'm aware of) comes from users hand writing either intrinsics...

@ngzhian > (Going to repeat some stuff here to make sure I understand things correctly.) I think your summary is exactly right! > ... [t]his could be an unexpected performance...

> Interesting thought. We can take this chance to get scalar fma and scalar qfma in as well. It's not a lot more work, represents a bit of scope creep,...

@zeux your "set" and "list" non-determinism should be reversed (although I should be clear that this is terminology I made up for this discussion!). Wasm's existing non-determinism is "set" non-determism...